4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * This file contains the top half of the zfs directory structure
29 * implementation. The bottom half is in zap_leaf.c.
31 * The zdir is an extendable hash data structure. There is a table of
32 * pointers to buckets (zap_t->zd_data->zd_leafs). The buckets are
33 * each a constant size and hold a variable number of directory entries.
34 * The buckets (aka "leaf nodes") are implemented in zap_leaf.c.
36 * The pointer table holds a power of 2 number of pointers.
37 * (1<<zap_t->zd_data->zd_phys->zd_prefix_len). The bucket pointed to
38 * by the pointer at index i in the table holds entries whose hash value
39 * has a zd_prefix_len - bit prefix
44 #include <sys/zfs_context.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/fs/zfs.h>
48 #include <sys/refcount.h>
49 #include <sys/zap_impl.h>
50 #include <sys/zap_leaf.h>
52 int fzap_default_block_shift = 14; /* 16k blocksize */
54 extern inline zap_phys_t *zap_f_phys(zap_t *zap);
56 static uint64_t zap_allocate_blocks(zap_t *zap, int nblocks);
59 fzap_byteswap(void *vbuf, size_t size)
63 block_type = *(uint64_t *)vbuf;
65 if (block_type == ZBT_LEAF || block_type == BSWAP_64(ZBT_LEAF))
66 zap_leaf_byteswap(vbuf, size);
68 /* it's a ptrtbl block */
69 byteswap_uint64_array(vbuf, size);
74 fzap_upgrade(zap_t *zap, dmu_tx_t *tx, zap_flags_t flags)
81 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
82 zap->zap_ismicro = FALSE;
84 zap->zap_dbu.dbu_evict_func = zap_evict;
86 mutex_init(&zap->zap_f.zap_num_entries_mtx, 0, 0, 0);
87 zap->zap_f.zap_block_shift = highbit64(zap->zap_dbuf->db_size) - 1;
91 * explicitly zero it since it might be coming from an
92 * initialized microzap
94 bzero(zap->zap_dbuf->db_data, zap->zap_dbuf->db_size);
95 zp->zap_block_type = ZBT_HEADER;
96 zp->zap_magic = ZAP_MAGIC;
98 zp->zap_ptrtbl.zt_shift = ZAP_EMBEDDED_PTRTBL_SHIFT(zap);
100 zp->zap_freeblk = 2; /* block 1 will be the first leaf */
101 zp->zap_num_leafs = 1;
102 zp->zap_num_entries = 0;
103 zp->zap_salt = zap->zap_salt;
104 zp->zap_normflags = zap->zap_normflags;
105 zp->zap_flags = flags;
107 /* block 1 will be the first leaf */
108 for (i = 0; i < (1<<zp->zap_ptrtbl.zt_shift); i++)
109 ZAP_EMBEDDED_PTRTBL_ENT(zap, i) = 1;
112 * set up block 1 - the first leaf
114 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
115 1<<FZAP_BLOCK_SHIFT(zap), FTAG, &db, DMU_READ_NO_PREFETCH));
116 dmu_buf_will_dirty(db, tx);
118 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
121 zap_leaf_init(l, zp->zap_normflags != 0);
123 kmem_free(l, sizeof (zap_leaf_t));
124 dmu_buf_rele(db, FTAG);
128 zap_tryupgradedir(zap_t *zap, dmu_tx_t *tx)
130 if (RW_WRITE_HELD(&zap->zap_rwlock))
132 if (rw_tryupgrade(&zap->zap_rwlock)) {
133 dmu_buf_will_dirty(zap->zap_dbuf, tx);
140 * Generic routines for dealing with the pointer & cookie tables.
144 zap_table_grow(zap_t *zap, zap_table_phys_t *tbl,
145 void (*transfer_func)(const uint64_t *src, uint64_t *dst, int n),
149 dmu_buf_t *db_old, *db_new;
151 int bs = FZAP_BLOCK_SHIFT(zap);
152 int hepb = 1<<(bs-4);
153 /* hepb = half the number of entries in a block */
155 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
156 ASSERT(tbl->zt_blk != 0);
157 ASSERT(tbl->zt_numblks > 0);
159 if (tbl->zt_nextblk != 0) {
160 newblk = tbl->zt_nextblk;
162 newblk = zap_allocate_blocks(zap, tbl->zt_numblks * 2);
163 tbl->zt_nextblk = newblk;
164 ASSERT0(tbl->zt_blks_copied);
165 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
166 tbl->zt_blk << bs, tbl->zt_numblks << bs,
167 ZIO_PRIORITY_SYNC_READ);
171 * Copy the ptrtbl from the old to new location.
174 b = tbl->zt_blks_copied;
175 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
176 (tbl->zt_blk + b) << bs, FTAG, &db_old, DMU_READ_NO_PREFETCH);
180 /* first half of entries in old[b] go to new[2*b+0] */
181 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
182 (newblk + 2*b+0) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
183 dmu_buf_will_dirty(db_new, tx);
184 transfer_func(db_old->db_data, db_new->db_data, hepb);
185 dmu_buf_rele(db_new, FTAG);
187 /* second half of entries in old[b] go to new[2*b+1] */
188 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
189 (newblk + 2*b+1) << bs, FTAG, &db_new, DMU_READ_NO_PREFETCH));
190 dmu_buf_will_dirty(db_new, tx);
191 transfer_func((uint64_t *)db_old->db_data + hepb,
192 db_new->db_data, hepb);
193 dmu_buf_rele(db_new, FTAG);
195 dmu_buf_rele(db_old, FTAG);
197 tbl->zt_blks_copied++;
199 dprintf("copied block %llu of %llu\n",
200 tbl->zt_blks_copied, tbl->zt_numblks);
202 if (tbl->zt_blks_copied == tbl->zt_numblks) {
203 (void) dmu_free_range(zap->zap_objset, zap->zap_object,
204 tbl->zt_blk << bs, tbl->zt_numblks << bs, tx);
206 tbl->zt_blk = newblk;
207 tbl->zt_numblks *= 2;
210 tbl->zt_blks_copied = 0;
212 dprintf("finished; numblocks now %llu (%lluk entries)\n",
213 tbl->zt_numblks, 1<<(tbl->zt_shift-10));
220 zap_table_store(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t val,
225 int bs = FZAP_BLOCK_SHIFT(zap);
228 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
229 ASSERT(tbl->zt_blk != 0);
231 dprintf("storing %llx at index %llx\n", val, idx);
234 off = idx & ((1<<(bs-3))-1);
236 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
237 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
240 dmu_buf_will_dirty(db, tx);
242 if (tbl->zt_nextblk != 0) {
243 uint64_t idx2 = idx * 2;
244 uint64_t blk2 = idx2 >> (bs-3);
245 uint64_t off2 = idx2 & ((1<<(bs-3))-1);
248 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
249 (tbl->zt_nextblk + blk2) << bs, FTAG, &db2,
250 DMU_READ_NO_PREFETCH);
252 dmu_buf_rele(db, FTAG);
255 dmu_buf_will_dirty(db2, tx);
256 ((uint64_t *)db2->db_data)[off2] = val;
257 ((uint64_t *)db2->db_data)[off2+1] = val;
258 dmu_buf_rele(db2, FTAG);
261 ((uint64_t *)db->db_data)[off] = val;
262 dmu_buf_rele(db, FTAG);
268 zap_table_load(zap_t *zap, zap_table_phys_t *tbl, uint64_t idx, uint64_t *valp)
273 int bs = FZAP_BLOCK_SHIFT(zap);
275 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
278 off = idx & ((1<<(bs-3))-1);
280 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
281 (tbl->zt_blk + blk) << bs, FTAG, &db, DMU_READ_NO_PREFETCH);
284 *valp = ((uint64_t *)db->db_data)[off];
285 dmu_buf_rele(db, FTAG);
287 if (tbl->zt_nextblk != 0) {
289 * read the nextblk for the sake of i/o error checking,
290 * so that zap_table_load() will catch errors for
293 blk = (idx*2) >> (bs-3);
295 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
296 (tbl->zt_nextblk + blk) << bs, FTAG, &db,
297 DMU_READ_NO_PREFETCH);
299 dmu_buf_rele(db, FTAG);
305 * Routines for growing the ptrtbl.
309 zap_ptrtbl_transfer(const uint64_t *src, uint64_t *dst, int n)
312 for (i = 0; i < n; i++) {
313 uint64_t lb = src[i];
320 zap_grow_ptrtbl(zap_t *zap, dmu_tx_t *tx)
323 * The pointer table should never use more hash bits than we
324 * have (otherwise we'd be using useless zero bits to index it).
325 * If we are within 2 bits of running out, stop growing, since
326 * this is already an aberrant condition.
328 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift >= zap_hashbits(zap) - 2)
329 return (SET_ERROR(ENOSPC));
331 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
333 * We are outgrowing the "embedded" ptrtbl (the one
334 * stored in the header block). Give it its own entire
335 * block, which will double the size of the ptrtbl.
341 ASSERT3U(zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
342 ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
343 ASSERT0(zap_f_phys(zap)->zap_ptrtbl.zt_blk);
345 newblk = zap_allocate_blocks(zap, 1);
346 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
347 newblk << FZAP_BLOCK_SHIFT(zap), FTAG, &db_new,
348 DMU_READ_NO_PREFETCH);
351 dmu_buf_will_dirty(db_new, tx);
352 zap_ptrtbl_transfer(&ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
353 db_new->db_data, 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap));
354 dmu_buf_rele(db_new, FTAG);
356 zap_f_phys(zap)->zap_ptrtbl.zt_blk = newblk;
357 zap_f_phys(zap)->zap_ptrtbl.zt_numblks = 1;
358 zap_f_phys(zap)->zap_ptrtbl.zt_shift++;
360 ASSERT3U(1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift, ==,
361 zap_f_phys(zap)->zap_ptrtbl.zt_numblks <<
362 (FZAP_BLOCK_SHIFT(zap)-3));
366 return (zap_table_grow(zap, &zap_f_phys(zap)->zap_ptrtbl,
367 zap_ptrtbl_transfer, tx));
372 zap_increment_num_entries(zap_t *zap, int delta, dmu_tx_t *tx)
374 dmu_buf_will_dirty(zap->zap_dbuf, tx);
375 mutex_enter(&zap->zap_f.zap_num_entries_mtx);
376 ASSERT(delta > 0 || zap_f_phys(zap)->zap_num_entries >= -delta);
377 zap_f_phys(zap)->zap_num_entries += delta;
378 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
382 zap_allocate_blocks(zap_t *zap, int nblocks)
385 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
386 newblk = zap_f_phys(zap)->zap_freeblk;
387 zap_f_phys(zap)->zap_freeblk += nblocks;
392 zap_leaf_pageout(void *dbu)
396 rw_destroy(&l->l_rwlock);
397 kmem_free(l, sizeof (zap_leaf_t));
401 zap_create_leaf(zap_t *zap, dmu_tx_t *tx)
404 zap_leaf_t *l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
406 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
408 rw_init(&l->l_rwlock, 0, 0, 0);
409 rw_enter(&l->l_rwlock, RW_WRITER);
410 l->l_blkid = zap_allocate_blocks(zap, 1);
413 VERIFY(0 == dmu_buf_hold(zap->zap_objset, zap->zap_object,
414 l->l_blkid << FZAP_BLOCK_SHIFT(zap), NULL, &l->l_dbuf,
415 DMU_READ_NO_PREFETCH));
416 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
417 winner = dmu_buf_set_user(l->l_dbuf, &l->l_dbu);
418 ASSERT(winner == NULL);
419 dmu_buf_will_dirty(l->l_dbuf, tx);
421 zap_leaf_init(l, zap->zap_normflags != 0);
423 zap_f_phys(zap)->zap_num_leafs++;
429 fzap_count(zap_t *zap, uint64_t *count)
431 ASSERT(!zap->zap_ismicro);
432 mutex_enter(&zap->zap_f.zap_num_entries_mtx); /* unnecessary */
433 *count = zap_f_phys(zap)->zap_num_entries;
434 mutex_exit(&zap->zap_f.zap_num_entries_mtx);
439 * Routines for obtaining zap_leaf_t's
443 zap_put_leaf(zap_leaf_t *l)
445 rw_exit(&l->l_rwlock);
446 dmu_buf_rele(l->l_dbuf, NULL);
450 zap_open_leaf(uint64_t blkid, dmu_buf_t *db)
452 zap_leaf_t *l, *winner;
456 l = kmem_zalloc(sizeof (zap_leaf_t), KM_SLEEP);
457 rw_init(&l->l_rwlock, 0, 0, 0);
458 rw_enter(&l->l_rwlock, RW_WRITER);
460 l->l_bs = highbit64(db->db_size) - 1;
463 dmu_buf_init_user(&l->l_dbu, zap_leaf_pageout, &l->l_dbuf);
464 winner = dmu_buf_set_user(db, &l->l_dbu);
466 rw_exit(&l->l_rwlock);
467 if (winner != NULL) {
468 /* someone else set it first */
469 zap_leaf_pageout(&l->l_dbu);
474 * lhr_pad was previously used for the next leaf in the leaf
475 * chain. There should be no chained leafs (as we have removed
478 ASSERT0(zap_leaf_phys(l)->l_hdr.lh_pad1);
481 * There should be more hash entries than there can be
482 * chunks to put in the hash table
484 ASSERT3U(ZAP_LEAF_HASH_NUMENTRIES(l), >, ZAP_LEAF_NUMCHUNKS(l) / 3);
486 /* The chunks should begin at the end of the hash table */
487 ASSERT3P(&ZAP_LEAF_CHUNK(l, 0), ==,
488 &zap_leaf_phys(l)->l_hash[ZAP_LEAF_HASH_NUMENTRIES(l)]);
490 /* The chunks should end at the end of the block */
491 ASSERT3U((uintptr_t)&ZAP_LEAF_CHUNK(l, ZAP_LEAF_NUMCHUNKS(l)) -
492 (uintptr_t)zap_leaf_phys(l), ==, l->l_dbuf->db_size);
498 zap_get_leaf_byblk(zap_t *zap, uint64_t blkid, dmu_tx_t *tx, krw_t lt,
503 int bs = FZAP_BLOCK_SHIFT(zap);
506 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
508 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
509 blkid << bs, NULL, &db, DMU_READ_NO_PREFETCH);
513 ASSERT3U(db->db_object, ==, zap->zap_object);
514 ASSERT3U(db->db_offset, ==, blkid << bs);
515 ASSERT3U(db->db_size, ==, 1 << bs);
518 l = dmu_buf_get_user(db);
521 l = zap_open_leaf(blkid, db);
523 rw_enter(&l->l_rwlock, lt);
525 * Must lock before dirtying, otherwise zap_leaf_phys(l) could change,
526 * causing ASSERT below to fail.
529 dmu_buf_will_dirty(db, tx);
530 ASSERT3U(l->l_blkid, ==, blkid);
531 ASSERT3P(l->l_dbuf, ==, db);
532 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_block_type, ==, ZBT_LEAF);
533 ASSERT3U(zap_leaf_phys(l)->l_hdr.lh_magic, ==, ZAP_LEAF_MAGIC);
540 zap_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t *valp)
542 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
544 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
546 (1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift));
547 *valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
550 return (zap_table_load(zap, &zap_f_phys(zap)->zap_ptrtbl,
556 zap_set_idx_to_blk(zap_t *zap, uint64_t idx, uint64_t blk, dmu_tx_t *tx)
559 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
561 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0) {
562 ZAP_EMBEDDED_PTRTBL_ENT(zap, idx) = blk;
565 return (zap_table_store(zap, &zap_f_phys(zap)->zap_ptrtbl,
571 zap_deref_leaf(zap_t *zap, uint64_t h, dmu_tx_t *tx, krw_t lt, zap_leaf_t **lp)
576 ASSERT(zap->zap_dbuf == NULL ||
577 zap_f_phys(zap) == zap->zap_dbuf->db_data);
578 ASSERT3U(zap_f_phys(zap)->zap_magic, ==, ZAP_MAGIC);
579 idx = ZAP_HASH_IDX(h, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
580 err = zap_idx_to_blk(zap, idx, &blk);
583 err = zap_get_leaf_byblk(zap, blk, tx, lt, lp);
586 ZAP_HASH_IDX(h, zap_leaf_phys(*lp)->l_hdr.lh_prefix_len) ==
587 zap_leaf_phys(*lp)->l_hdr.lh_prefix);
592 zap_expand_leaf(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx, zap_leaf_t **lp)
594 zap_t *zap = zn->zn_zap;
595 uint64_t hash = zn->zn_hash;
597 int prefix_diff, i, err;
599 int old_prefix_len = zap_leaf_phys(l)->l_hdr.lh_prefix_len;
601 ASSERT3U(old_prefix_len, <=, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
602 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
604 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
605 zap_leaf_phys(l)->l_hdr.lh_prefix);
607 if (zap_tryupgradedir(zap, tx) == 0 ||
608 old_prefix_len == zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
609 /* We failed to upgrade, or need to grow the pointer table */
610 objset_t *os = zap->zap_objset;
611 uint64_t object = zap->zap_object;
615 err = zap_lockdir(os, object, tx, RW_WRITER,
616 FALSE, FALSE, &zn->zn_zap);
620 ASSERT(!zap->zap_ismicro);
622 while (old_prefix_len ==
623 zap_f_phys(zap)->zap_ptrtbl.zt_shift) {
624 err = zap_grow_ptrtbl(zap, tx);
629 err = zap_deref_leaf(zap, hash, tx, RW_WRITER, &l);
633 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len != old_prefix_len) {
634 /* it split while our locks were down */
639 ASSERT(RW_WRITE_HELD(&zap->zap_rwlock));
640 ASSERT3U(old_prefix_len, <, zap_f_phys(zap)->zap_ptrtbl.zt_shift);
641 ASSERT3U(ZAP_HASH_IDX(hash, old_prefix_len), ==,
642 zap_leaf_phys(l)->l_hdr.lh_prefix);
644 prefix_diff = zap_f_phys(zap)->zap_ptrtbl.zt_shift -
645 (old_prefix_len + 1);
646 sibling = (ZAP_HASH_IDX(hash, old_prefix_len + 1) | 1) << prefix_diff;
648 /* check for i/o errors before doing zap_leaf_split */
649 for (i = 0; i < (1ULL<<prefix_diff); i++) {
651 err = zap_idx_to_blk(zap, sibling+i, &blk);
654 ASSERT3U(blk, ==, l->l_blkid);
657 nl = zap_create_leaf(zap, tx);
658 zap_leaf_split(l, nl, zap->zap_normflags != 0);
660 /* set sibling pointers */
661 for (i = 0; i < (1ULL << prefix_diff); i++) {
662 err = zap_set_idx_to_blk(zap, sibling+i, nl->l_blkid, tx);
663 ASSERT0(err); /* we checked for i/o errors above */
666 if (hash & (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len))) {
667 /* we want the sibling */
679 zap_put_leaf_maybe_grow_ptrtbl(zap_name_t *zn, zap_leaf_t *l, dmu_tx_t *tx)
681 zap_t *zap = zn->zn_zap;
682 int shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
683 int leaffull = (zap_leaf_phys(l)->l_hdr.lh_prefix_len == shift &&
684 zap_leaf_phys(l)->l_hdr.lh_nfree < ZAP_LEAF_LOW_WATER);
688 if (leaffull || zap_f_phys(zap)->zap_ptrtbl.zt_nextblk) {
692 * We are in the middle of growing the pointer table, or
693 * this leaf will soon make us grow it.
695 if (zap_tryupgradedir(zap, tx) == 0) {
696 objset_t *os = zap->zap_objset;
697 uint64_t zapobj = zap->zap_object;
700 err = zap_lockdir(os, zapobj, tx,
701 RW_WRITER, FALSE, FALSE, &zn->zn_zap);
707 /* could have finished growing while our locks were down */
708 if (zap_f_phys(zap)->zap_ptrtbl.zt_shift == shift)
709 (void) zap_grow_ptrtbl(zap, tx);
714 fzap_checkname(zap_name_t *zn)
716 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
717 return (SET_ERROR(ENAMETOOLONG));
722 fzap_checksize(uint64_t integer_size, uint64_t num_integers)
724 /* Only integer sizes supported by C */
725 switch (integer_size) {
732 return (SET_ERROR(EINVAL));
735 if (integer_size * num_integers > ZAP_MAXVALUELEN)
742 fzap_check(zap_name_t *zn, uint64_t integer_size, uint64_t num_integers)
746 if ((err = fzap_checkname(zn)) != 0)
748 return (fzap_checksize(integer_size, num_integers));
752 * Routines for manipulating attributes.
755 fzap_lookup(zap_name_t *zn,
756 uint64_t integer_size, uint64_t num_integers, void *buf,
757 char *realname, int rn_len, boolean_t *ncp)
761 zap_entry_handle_t zeh;
763 if ((err = fzap_checkname(zn)) != 0)
766 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
769 err = zap_leaf_lookup(l, zn, &zeh);
771 if ((err = fzap_checksize(integer_size, num_integers)) != 0) {
776 err = zap_entry_read(&zeh, integer_size, num_integers, buf);
777 (void) zap_entry_read_name(zn->zn_zap, &zeh, rn_len, realname);
779 *ncp = zap_entry_normalization_conflict(&zeh,
780 zn, NULL, zn->zn_zap);
789 fzap_add_cd(zap_name_t *zn,
790 uint64_t integer_size, uint64_t num_integers,
791 const void *val, uint32_t cd, dmu_tx_t *tx)
795 zap_entry_handle_t zeh;
796 zap_t *zap = zn->zn_zap;
798 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
799 ASSERT(!zap->zap_ismicro);
800 ASSERT(fzap_check(zn, integer_size, num_integers) == 0);
802 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
806 err = zap_leaf_lookup(l, zn, &zeh);
808 err = SET_ERROR(EEXIST);
814 err = zap_entry_create(l, zn, cd,
815 integer_size, num_integers, val, &zeh);
818 zap_increment_num_entries(zap, 1, tx);
819 } else if (err == EAGAIN) {
820 err = zap_expand_leaf(zn, l, tx, &l);
821 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
828 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
833 fzap_add(zap_name_t *zn,
834 uint64_t integer_size, uint64_t num_integers,
835 const void *val, dmu_tx_t *tx)
837 int err = fzap_check(zn, integer_size, num_integers);
841 return (fzap_add_cd(zn, integer_size, num_integers,
842 val, ZAP_NEED_CD, tx));
846 fzap_update(zap_name_t *zn,
847 int integer_size, uint64_t num_integers, const void *val, dmu_tx_t *tx)
851 zap_entry_handle_t zeh;
852 zap_t *zap = zn->zn_zap;
854 ASSERT(RW_LOCK_HELD(&zap->zap_rwlock));
855 err = fzap_check(zn, integer_size, num_integers);
859 err = zap_deref_leaf(zap, zn->zn_hash, tx, RW_WRITER, &l);
863 err = zap_leaf_lookup(l, zn, &zeh);
864 create = (err == ENOENT);
865 ASSERT(err == 0 || err == ENOENT);
868 err = zap_entry_create(l, zn, ZAP_NEED_CD,
869 integer_size, num_integers, val, &zeh);
871 zap_increment_num_entries(zap, 1, tx);
873 err = zap_entry_update(&zeh, integer_size, num_integers, val);
877 err = zap_expand_leaf(zn, l, tx, &l);
878 zap = zn->zn_zap; /* zap_expand_leaf() may change zap */
884 zap_put_leaf_maybe_grow_ptrtbl(zn, l, tx);
889 fzap_length(zap_name_t *zn,
890 uint64_t *integer_size, uint64_t *num_integers)
894 zap_entry_handle_t zeh;
896 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, NULL, RW_READER, &l);
899 err = zap_leaf_lookup(l, zn, &zeh);
904 *integer_size = zeh.zeh_integer_size;
906 *num_integers = zeh.zeh_num_integers;
913 fzap_remove(zap_name_t *zn, dmu_tx_t *tx)
917 zap_entry_handle_t zeh;
919 err = zap_deref_leaf(zn->zn_zap, zn->zn_hash, tx, RW_WRITER, &l);
922 err = zap_leaf_lookup(l, zn, &zeh);
924 zap_entry_remove(&zeh);
925 zap_increment_num_entries(zn->zn_zap, -1, tx);
932 fzap_prefetch(zap_name_t *zn)
935 zap_t *zap = zn->zn_zap;
938 idx = ZAP_HASH_IDX(zn->zn_hash,
939 zap_f_phys(zap)->zap_ptrtbl.zt_shift);
940 if (zap_idx_to_blk(zap, idx, &blk) != 0)
942 bs = FZAP_BLOCK_SHIFT(zap);
943 dmu_prefetch(zap->zap_objset, zap->zap_object, 0, blk << bs, 1 << bs,
944 ZIO_PRIORITY_SYNC_READ);
948 * Helper functions for consumers.
952 zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
953 const char *name, dmu_tx_t *tx)
957 VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
958 VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
965 zap_value_search(objset_t *os, uint64_t zapobj, uint64_t value, uint64_t mask,
975 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
976 for (zap_cursor_init(&zc, os, zapobj);
977 (err = zap_cursor_retrieve(&zc, za)) == 0;
978 zap_cursor_advance(&zc)) {
979 if ((za->za_first_integer & mask) == (value & mask)) {
980 (void) strcpy(name, za->za_name);
984 zap_cursor_fini(&zc);
985 kmem_free(za, sizeof (zap_attribute_t));
990 zap_join(objset_t *os, uint64_t fromobj, uint64_t intoobj, dmu_tx_t *tx)
997 for (zap_cursor_init(&zc, os, fromobj);
998 zap_cursor_retrieve(&zc, &za) == 0;
999 (void) zap_cursor_advance(&zc)) {
1000 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1001 err = SET_ERROR(EINVAL);
1004 err = zap_add(os, intoobj, za.za_name,
1005 8, 1, &za.za_first_integer, tx);
1009 zap_cursor_fini(&zc);
1014 zap_join_key(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1015 uint64_t value, dmu_tx_t *tx)
1022 for (zap_cursor_init(&zc, os, fromobj);
1023 zap_cursor_retrieve(&zc, &za) == 0;
1024 (void) zap_cursor_advance(&zc)) {
1025 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1026 err = SET_ERROR(EINVAL);
1029 err = zap_add(os, intoobj, za.za_name,
1034 zap_cursor_fini(&zc);
1039 zap_join_increment(objset_t *os, uint64_t fromobj, uint64_t intoobj,
1047 for (zap_cursor_init(&zc, os, fromobj);
1048 zap_cursor_retrieve(&zc, &za) == 0;
1049 (void) zap_cursor_advance(&zc)) {
1052 if (za.za_integer_length != 8 || za.za_num_integers != 1) {
1053 err = SET_ERROR(EINVAL);
1057 err = zap_lookup(os, intoobj, za.za_name, 8, 1, &delta);
1058 if (err != 0 && err != ENOENT)
1060 delta += za.za_first_integer;
1061 err = zap_update(os, intoobj, za.za_name, 8, 1, &delta, tx);
1065 zap_cursor_fini(&zc);
1070 zap_add_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1074 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1075 return (zap_add(os, obj, name, 8, 1, &value, tx));
1079 zap_remove_int(objset_t *os, uint64_t obj, uint64_t value, dmu_tx_t *tx)
1083 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1084 return (zap_remove(os, obj, name, tx));
1088 zap_lookup_int(objset_t *os, uint64_t obj, uint64_t value)
1092 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)value);
1093 return (zap_lookup(os, obj, name, 8, 1, &value));
1097 zap_add_int_key(objset_t *os, uint64_t obj,
1098 uint64_t key, uint64_t value, dmu_tx_t *tx)
1102 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1103 return (zap_add(os, obj, name, 8, 1, &value, tx));
1107 zap_update_int_key(objset_t *os, uint64_t obj,
1108 uint64_t key, uint64_t value, dmu_tx_t *tx)
1112 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1113 return (zap_update(os, obj, name, 8, 1, &value, tx));
1117 zap_lookup_int_key(objset_t *os, uint64_t obj, uint64_t key, uint64_t *valuep)
1121 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1122 return (zap_lookup(os, obj, name, 8, 1, valuep));
1126 zap_increment(objset_t *os, uint64_t obj, const char *name, int64_t delta,
1135 err = zap_lookup(os, obj, name, 8, 1, &value);
1136 if (err != 0 && err != ENOENT)
1140 err = zap_remove(os, obj, name, tx);
1142 err = zap_update(os, obj, name, 8, 1, &value, tx);
1147 zap_increment_int(objset_t *os, uint64_t obj, uint64_t key, int64_t delta,
1152 (void) snprintf(name, sizeof (name), "%llx", (longlong_t)key);
1153 return (zap_increment(os, obj, name, delta, tx));
1157 * Routines for iterating over the attributes.
1161 fzap_cursor_retrieve(zap_t *zap, zap_cursor_t *zc, zap_attribute_t *za)
1164 zap_entry_handle_t zeh;
1167 /* retrieve the next entry at or after zc_hash/zc_cd */
1168 /* if no entry, return ENOENT */
1171 (ZAP_HASH_IDX(zc->zc_hash,
1172 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix_len) !=
1173 zap_leaf_phys(zc->zc_leaf)->l_hdr.lh_prefix)) {
1174 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1175 zap_put_leaf(zc->zc_leaf);
1180 if (zc->zc_leaf == NULL) {
1181 err = zap_deref_leaf(zap, zc->zc_hash, NULL, RW_READER,
1186 rw_enter(&zc->zc_leaf->l_rwlock, RW_READER);
1190 err = zap_leaf_lookup_closest(l, zc->zc_hash, zc->zc_cd, &zeh);
1192 if (err == ENOENT) {
1194 (1ULL << (64 - zap_leaf_phys(l)->l_hdr.lh_prefix_len)) - 1;
1195 zc->zc_hash = (zc->zc_hash & ~nocare) + nocare + 1;
1197 if (zap_leaf_phys(l)->l_hdr.lh_prefix_len == 0 ||
1199 zc->zc_hash = -1ULL;
1201 zap_put_leaf(zc->zc_leaf);
1208 zc->zc_hash = zeh.zeh_hash;
1209 zc->zc_cd = zeh.zeh_cd;
1210 za->za_integer_length = zeh.zeh_integer_size;
1211 za->za_num_integers = zeh.zeh_num_integers;
1212 if (zeh.zeh_num_integers == 0) {
1213 za->za_first_integer = 0;
1215 err = zap_entry_read(&zeh, 8, 1, &za->za_first_integer);
1216 ASSERT(err == 0 || err == EOVERFLOW);
1218 err = zap_entry_read_name(zap, &zeh,
1219 sizeof (za->za_name), za->za_name);
1222 za->za_normalization_conflict =
1223 zap_entry_normalization_conflict(&zeh,
1224 NULL, za->za_name, zap);
1226 rw_exit(&zc->zc_leaf->l_rwlock);
1231 zap_stats_ptrtbl(zap_t *zap, uint64_t *tbl, int len, zap_stats_t *zs)
1234 uint64_t lastblk = 0;
1237 * NB: if a leaf has more pointers than an entire ptrtbl block
1238 * can hold, then it'll be accounted for more than once, since
1239 * we won't have lastblk.
1241 for (i = 0; i < len; i++) {
1244 if (tbl[i] == lastblk)
1248 err = zap_get_leaf_byblk(zap, tbl[i], NULL, RW_READER, &l);
1250 zap_leaf_stats(zap, l, zs);
1257 fzap_cursor_move_to_key(zap_cursor_t *zc, zap_name_t *zn)
1261 zap_entry_handle_t zeh;
1263 if (zn->zn_key_orig_numints * zn->zn_key_intlen > ZAP_MAXNAMELEN)
1264 return (SET_ERROR(ENAMETOOLONG));
1266 err = zap_deref_leaf(zc->zc_zap, zn->zn_hash, NULL, RW_READER, &l);
1270 err = zap_leaf_lookup(l, zn, &zeh);
1275 zc->zc_hash = zeh.zeh_hash;
1276 zc->zc_cd = zeh.zeh_cd;
1282 fzap_get_stats(zap_t *zap, zap_stats_t *zs)
1284 int bs = FZAP_BLOCK_SHIFT(zap);
1285 zs->zs_blocksize = 1ULL << bs;
1288 * Set zap_phys_t fields
1290 zs->zs_num_leafs = zap_f_phys(zap)->zap_num_leafs;
1291 zs->zs_num_entries = zap_f_phys(zap)->zap_num_entries;
1292 zs->zs_num_blocks = zap_f_phys(zap)->zap_freeblk;
1293 zs->zs_block_type = zap_f_phys(zap)->zap_block_type;
1294 zs->zs_magic = zap_f_phys(zap)->zap_magic;
1295 zs->zs_salt = zap_f_phys(zap)->zap_salt;
1298 * Set zap_ptrtbl fields
1300 zs->zs_ptrtbl_len = 1ULL << zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1301 zs->zs_ptrtbl_nextblk = zap_f_phys(zap)->zap_ptrtbl.zt_nextblk;
1302 zs->zs_ptrtbl_blks_copied =
1303 zap_f_phys(zap)->zap_ptrtbl.zt_blks_copied;
1304 zs->zs_ptrtbl_zt_blk = zap_f_phys(zap)->zap_ptrtbl.zt_blk;
1305 zs->zs_ptrtbl_zt_numblks = zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1306 zs->zs_ptrtbl_zt_shift = zap_f_phys(zap)->zap_ptrtbl.zt_shift;
1308 if (zap_f_phys(zap)->zap_ptrtbl.zt_numblks == 0) {
1309 /* the ptrtbl is entirely in the header block. */
1310 zap_stats_ptrtbl(zap, &ZAP_EMBEDDED_PTRTBL_ENT(zap, 0),
1311 1 << ZAP_EMBEDDED_PTRTBL_SHIFT(zap), zs);
1315 dmu_prefetch(zap->zap_objset, zap->zap_object, 0,
1316 zap_f_phys(zap)->zap_ptrtbl.zt_blk << bs,
1317 zap_f_phys(zap)->zap_ptrtbl.zt_numblks << bs,
1318 ZIO_PRIORITY_SYNC_READ);
1320 for (b = 0; b < zap_f_phys(zap)->zap_ptrtbl.zt_numblks;
1325 err = dmu_buf_hold(zap->zap_objset, zap->zap_object,
1326 (zap_f_phys(zap)->zap_ptrtbl.zt_blk + b) << bs,
1327 FTAG, &db, DMU_READ_NO_PREFETCH);
1329 zap_stats_ptrtbl(zap, db->db_data,
1331 dmu_buf_rele(db, FTAG);
1338 fzap_count_write(zap_name_t *zn, int add, uint64_t *towrite,
1339 uint64_t *tooverwrite)
1341 zap_t *zap = zn->zn_zap;
1346 * Account for the header block of the fatzap.
1348 if (!add && dmu_buf_freeable(zap->zap_dbuf)) {
1349 *tooverwrite += zap->zap_dbuf->db_size;
1351 *towrite += zap->zap_dbuf->db_size;
1355 * Account for the pointer table blocks.
1356 * If we are adding we need to account for the following cases :
1357 * - If the pointer table is embedded, this operation could force an
1358 * external pointer table.
1359 * - If this already has an external pointer table this operation
1360 * could extend the table.
1363 if (zap_f_phys(zap)->zap_ptrtbl.zt_blk == 0)
1364 *towrite += zap->zap_dbuf->db_size;
1366 *towrite += (zap->zap_dbuf->db_size * 3);
1370 * Now, check if the block containing leaf is freeable
1371 * and account accordingly.
1373 err = zap_deref_leaf(zap, zn->zn_hash, NULL, RW_READER, &l);
1378 if (!add && dmu_buf_freeable(l->l_dbuf)) {
1379 *tooverwrite += l->l_dbuf->db_size;
1382 * If this an add operation, the leaf block could split.
1383 * Hence, we need to account for an additional leaf block.
1385 *towrite += (add ? 2 : 1) * l->l_dbuf->db_size;